# ashvardanian/NumKong

SIMD-accelerated distances, dot products, matrix ops, geospatial & geometric kernels for 16 numeric types — from 6-bit floats to 64-bit complex — across x86, Arm, RISC-V, and WASM, with bindings for Python, Rust, C, C++, Swift, JS, and Go 📐

Repository: https://github.com/ashvardanian/NumKong
Canonical: https://ross.abutalabs.com/products/numkong
Homepage: https://ashvardanian.com/posts/numkong
Language: C
License: Apache-2.0
License Family: permissive
Topics: arm-neon, assembly, metrics, simd, information-retrieval, numpy, scipy, vector-search, blas, matrix-multiplication, rust, tensor, cpp, golang, javascript, swift
Last push: 2026-08-22T20:50:42+00:00

## Health v2 (maintenance only)
Score: 93/100 (v2, computed 2026-09-03T02:39:23.370411+00:00)
- activity 99, release rhythm 87, longevity 90
- inputs: {"age_days": 1268, "days_push": 11, "days_rel": 11, "gap_med": 1, "n_releases_24m": 106}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1877, forks 128 (observed 2026-08-28T04:05:47.796628+00:00)

## What it is
NumKong is a mixed-precision numeric and linear-algebra library with ~2,000 SIMD-accelerated kernels for dot products, distances, matrix multiplication, geospatial, and geometric computations across 15+ numeric types (from 4-bit integers and 6-bit floats to 128-bit complex). It targets x86, Arm, RISC-V, LoongArch, Power, and WebAssembly, with bindings for C, C++, Python, Rust, JavaScript, Go, and Swift.

## Use cases
- compute dot products and distances faster than numpy or scipy
- run mixed-precision matrix multiplication on cpu
- accelerate vector search similarity metrics with simd
- calculate geodesic distances faster than geopy
- score colbert maxsim late interaction without a gpu
- use low-precision int4 and float6 kernels in python
- run blask-like math in webassembly in the browser

## When to choose
- you need fast, numerically stable low-precision math (int8, fp16, bf16, fp6) on cpu
- you want a tiny, dependency-free BLAS-like library with multi-language bindings
- you need SIMD portability across x86, Arm, RISC-V, and WASM
- you need single-call low latency rather than graph-based throughput frameworks

## When to avoid
- you need GPU or distributed training workloads where PyTorch/JAX excel
- you need the full BLAS/LAPACK API surface (eigen decomposition, solvers)
- you rely on a mature ecosystem with long-term community support

## Facets
- artifact type: library
- maturity: active
- function: math, benchmarking, machine-learning, geospatial, search-engine
- domain: performance, machine-learning, data-science, developer-tools
- platform: python, rust, cpp, go, wasm, cross-platform
- tags: simd, blas, linear-algebra, mixed-precision, dot-product, matrix-multiplication, vector-search, arm-neon, amx, sme, risc-v, numpy, scipy, tensors, distance-metrics, algorithms, javascript, swift, gpu

## Member repositories
- ashvardanian/NumKong (main) score 93

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:05:47.796628+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T03:14:41.680635+00:00, confidence not recorded.
  - readme: https://github.com/ashvardanian/NumKong (fetched 2026-08-28T04:05:47.796628+00:00, sha 4b64402e5b1c)
  - homepage: https://ashvardanian.com/posts/numkong (fetched 2026-08-29T10:53:25.843533+00:00, sha c40d09ab4879)
- Data as of 2026-08-30T08:39:29.467469+00:00.